Figure3

Figure 3. Comparisons of average game score and number of victories in challenging scenarios for the manually tuned model (base), the product of reinforcement learning (RL), and randomly initialized models (random).